A Channel and Spatial Attention-Based ResNet Module for EEG-Based Schizophrenia Detection

As a severe mental illness that acutely impairs patients’ cognitive flexibility, verbal communication and motor coordination among other facets of their life, schizophrenia must be identified as early as possible. Schizophrenia is very complex in nature as its symptoms fairly coincide with that of other illnesses, thus making sophisticated diagnostic equipment a requirement to have a better understanding and possibly better treatment results. As a substitute for this, the EEG technology has become very significant, offering an economical and noninvasive way of determining the neuropsychological origins of SZ. This study seeks to improve the classification accuracy of individuals with schizophrenia by identifying a better feature set to describe electroencephalography (EEG) signals. We offer an Attention based Resnet101 model based on Recurrence Plots (RP) to classify EEG signals of schizophrenic (SZ) patients from that of Healthy Control (HC) patients. A nonlinear technique for examining the periodicity, chaos, and non-stationarity of time series is the recurrence plot. It works well for processing short time series data. The proposed method is trained on two EEG datasets, one from the Institute of Psychiatry and Neurology in Warsaw, Poland and the other from https://www.kaggle.com/broach/button-tone-sz. The model achieved an accuracy of 99.37% and 95% respectively on both datasets. Further, the proposed work is compared with seven SOTA techniques and achieved a better performance, proving its effectiveness in medical applications.